i am curious: why this instead of feeding your LLM an OpenAPI spec?